De-noising magnetotelluric data using variational mode decomposition combined with mathematical morphology filtering and wavelet thresholding

Abstract

De-noising magnetotelluric (MT) data has become a key research topic closely related to the application of the MT method itself. Variational mode decomposition (VMD), a newly developed non-recursive signal decomposition method, has been applied to the denoising of MT data. However, some high-frequency noise may remain after VMD denoising. Moreover, the useful low-frequency signal and large-scale noise cannot always be easily separated, resulting in the loss of useful low-frequency signal components during signal reconstruction. To overcome this problem, a new time series editing strategy is proposed, which combines wavelet thresholding (WT) and mathematical morphology filtering (MMF) for further processing of the VMD output. WT is used to remove the remaining high-frequency noise from the retained signal after VMD, while MMF is utilized to extract the useful low-frequency signal components from the signal rejected by VMD. Finally, the WT and MMF outputs are merged to obtain the final denoising results. The proposed method was evaluated using a simulated dataset and two real datasets collected in the Hami Basin of East Tianshan, China. The simulated data tests showed that the proposed method has superior denoising ability compared to other time-domain methods, while the real data study showed that the quality of MT responses obtained from the proposed method is superior to that from conventional methods, especially for the low-frequency band.